Neuromorphic Photonic Memory Devices Using Ultrafast, Non-Volatile Phase-Change Materials.

Abstract

The search for ultrafast photonic memory devices is inspired by the ever-increasing number of cloud-computing, supercomputing, and artificial-intelligence applications, together with the unique advantages of signal processing in the optical domain such as high speed, large bandwidth, and low energy consumption. By embracing silicon photonics with chalcogenide phase-change materials (PCMs), non-volatile integrated photonic memory is developed with promising potential in photonic integrated circuits and nanophotonic applications. While conventional PCMs suffer from slow crystallization speed, scandium-doped antimony telluride (SST) has been recently developed for ultrafast phase-change random-access memory applications. An ultrafast non-volatile photonic memory based on an SST thin film with a 2 ns write/erase speed is demonstrated, which is the fastest write/erase speed ever reported in integrated phase-change photonic devices. SST-based photonic memories exhibit multilevel capabilities and good stability at room temperature. By mapping the memory level to the biological synapse weight, an artificial neural network based on photonic memory devices is successfully established for image classification. Additionally, a reflective nanodisplay application using SST with optoelectronic modulation capabilities is demonstrated. Both the optical and electrical changes in SST during the phase transition and the fast-switching speed demonstrate their potential for use in photonic computing, neuromorphic computing, nanophotonics, and optoelectronic applications.